We decided to move the C++ bindings, Qt bindings and Python bindings out of the gpgme repository into separate repositories. This will likely be done after the next release of gpgme. Even though we'll keep ABI compatibility we will bump the major version to 2.
The separation of the three bindings from gpgme has been prepared in T7110: Distribute the GpgME bindings separately from GpgME.
| rM GPGME | |||
| rM569348ee8406 cpp: Add pkgconfig file for gpgmepp | |||
If the major version for QGpgME is bumped, shouldn't we at least remove the virtual base classes. Eg: delete FooJob and rename QGpgMEFooJob to FooJob. I did regret not doing this when i moved them out of libkleo since this architecture no longer makes sense in the standalone libnrary and technically the virtual bases make it nearly impossible to maintain ABI stability when adding functions. The reason for those was only because libkleo had that idea of different backends namely gpgme and chiasmus. But a Library called QGpgME should never provide another backend then GPGME IMO.
So no behavioural change at all, just something to make future ABI compat easier.
Werner wrote:
We will bump the gpgme core version to 2.0 to indicate this split despite that there will be non-ABI/API incompatibility.
I read this as bumping the version-number e.g. from 1.24.5 to 2.0.0 without e.g. bumping the soname or changing the api_version as specified in the .pc file. (FWIW I think that is a great plan.)
Ah, ok I understood it as "we will change the soname for other reasons e.g. so that both versions are co installable but we will not break ABI". And I would prefer the break for qgpgme at least because of the mentioned problem not because I don't care about ABI stability but because I do and this is a big problem which only exists, because I didn't do it with the last repo move. There is no technical reason anymore for the abstract base classes.
We'll do this with QGpgME 3. And it's easy to add new functions by using the NVI pattern and, if needed, virtual functions in the attached private classes. I've been using this technique for quite some time now.
Right. This is just to avoid questions from users where they can find gpgme-2.0 matching the gpgmepp-2.0. No ABI changed planned.